Date: Sat 4 Jun 1988 01:06-EDT From: AIList Moderator Nick Papadakis Reply-To: AIList@AI.AI.MIT.EDU Us-Mail: MIT Mail Stop 38-390, Cambridge MA 02139 Phone: (617) 253-2737 Subject: AIList Digest V7 #18 To: AIList@AI.AI.MIT.EDU Status: R AIList Digest Saturday, 4 Jun 1988 Volume 7 : Issue 18 Today's Topics: Symbolics stock randomness Fredkin chess tournament results & comment Queries - AI in weather forecasting frame based languages parallel inference Response to: connectionist medical expert systems ---------------------------------------------------------------------- Date: 2 Jun 88 21:19:16 GMT From: heldeib@gmu90x.gmu.edu (heldeib) Subject: Re: Symbolics stock I read in a recent article that many of the AI companies including Symbolics, of course LMI, Xerox, and software oriented companies were losing money. I can't find the article at the moment but I recall that Symbolics lost a great deal of money and that TI was the only Lisp machine producer that was semi-decent financially. I'll try to find the article and post the statistics, but I guess this explains the decline in Symbolics Stock. I wonder how the others are doing ! If anyone saw that article plese post it ! I can't remember the source but it was either an IEEE magazine, AI-Expert, or perhaps Digital Review ! Hany K. Eldeib Department of Electrical and Computer Engineering George Mason University Fairfax, VA 22030 UUCP: uunet!pyrdc!gmu90x!heldeib Bitnet: heldeib@gmuvax Internet: heldeib@gmuvax.gmu.edu ------------------------------ Date: Fri, 3 Jun 88 10:15:30 EDT From: csrobe@icase.arpa (Charles S. Roberson) Subject: Re: randomness In AIList Digest, Thursday, 2 Jun 1988, V7 #15, Antti Ylikoski writes: >From: Antti Ylikoski > >In AIList Digest V7 #4, Barry Kort writes: > >>If I wanted to give my von Neumann machine a *true* random number >>generator, I would connect it to an A/D converter driven by thermal >>noise (i.e. a toasty resister). > >I recall that a Zener diode is a good source of noise (but cannot remember >the spectrum it gives). > >It could be a good idea to utilize a Zener / A-D converter random number >generator in Monte Carlo simulations. First off, I know nothing about thermal noise or Zener diodes, however I think I know a little about random number generation ("A little knowledge *is* a dangerous thing"). Second, Generally when one wants to generate random numbers, they wish to draw from a known distribution (Uniform, Normal, Poisson, Exponential, etc.). This generally means either they need a different generator for each distribution or they have a single generator than can be used to build the other distributions. (e.g. A Linear Congruential (Lehmer) Generator is a good choice since it generates Uniform(0.0,0.1) quite easily). Third, how random would a Zener / A-D (ZAD) random number generator be? There are certain characteristics that are considered to be necessary for *true* randomness Is the generator full-period? (i.e. does it generate all possible values before repeating a value?) How random does it appear to the eye? (a rather subjective test but it can be useful.) How reliable is the generator? What is the longest run that one can expect in a sequence of random numbers, and how many runs are there? (e.g 78 1 2 3 4 156) contains a run of length 4. How would it fair under Knuths' Spectral Test? How well are the numbers distributed in n-space? My guess is that using a Zener / A-D converter as a random number source will prove to be a biased system in which certain numbers are much more likely than others. It may very well be the case that the ZAD generator generates a very nice gauss(a,b) or some other distribution, but how confident are you of it working correctly? On close inspection one is likely to find that a ZAD generator would be as bad as most other hardware dependant (bit and clock watching, etc.) methods. There is a paper coming out in one of the next few issues of Communications of the ACM by S. K. Park and K. Miller that has a very thorough discussion of random number generation (including history, red herrings (IBM's RANU), and a proposed minimal standard). For anyone who wants to give their von Neumann machine a (portable) "*true* random number generator", I think they should read this article. I also think anyone who reads it will be amazed at the number of textbook authors who are still advocating substandard or just plain *bad* generators! (Some of those books are AI books!) I did not see Barry Korts original posting, so I don't know the context of his message. However, I don't think you can do significantly better than a Lehmer generator for a von Neumann machine -- its elegant in its mathematical simplicity, it has been tested over and over (it has years of mathematics behind it), and it performs quite well on Knuth's Spectral Test. I appologize for waxing on about random number generators but valid research can only be done with validated tools and a bad random number generator can destroy any simulation. Know your tools. -chip +-------------------------------------------------------------------------+ |Charles S. Roberson ARPANET: csrobe@icase.arpa | |ICASE, MS 132C BITNET: $csrobe@wmmvs.bitnet | |NASA Langley Rsch. Ctr. UUCP: ...!uunet!pyrdc!gmu90x!wmcs!csrobe| |Hampton, VA 23665-5225 Phone: (804) 865-4090 +-------------------------------------------------------------------------+ ------------------------------ Date: 3 Jun 88 18:03:02 GMT From: sun1.uucp!cracraft@jpl-elroy.arpa (Stuart Cracraft) Subject: Fredkin chess tournament results & comment Article 1178 of rec.games.chess: From: fhh@unh.cs.cmu.edu (Feng-Hsiung Hsu) Newsgroups: rec.games.chess Subject: Computers in Fredkin Masters Open Keywords: Chess, Computers Message-ID: <1828@pt.cs.cmu.edu> Date: 2 Jun 88 15:50:07 GMT Sender: netnews@pt.cs.cmu.edu Organization: Carnegie-Mellon University, CS/RI Lines: 61 Each year in the last few years, the Fredkin Foundation sponsored a chess event designed to promote computer chess research. Traditionally, the reigning ACM North American Computer Chess Champion is invited along with possibly some of the stronger programs at the time. This year, the Fredkin Masters Open was held from May 28 to May 30 on CMU campus. About 30 masters participated. The computer opposition includes ChipTest, the reigning ACM Champion, Deep Thought 0.01 (0.01 stands for single processor, successor to ChipTest), Hitech (1985 ACM Champion) and BP (a Compaq 386). Phoenix (number 3 finisher in ACM) and Lachex (number 4 finisher in ACM) did not participate because of problems in obtaining computing time. Alexander Ivanov (FIDE 2415), a recent Soviet emigre, won the event by scoring 5 out of 6 and received the $1,200 first prize. Deep Thought tied for 2nd with 2 masters, scoring 4.5 out of 6. ChipTest tied for 5th scoring 4 out of 6. Hitech scored 3.5 out of 6. BP scored 3 out of 6. Based on the 3-month old ratings of the opponents, both Deep Thought and ChipTest should obtain provisional ratings above 2500. Deep Thought beat a 2339, drew a 2292, won against a 2299 and a 2389, lost to Ivanov, and won against Vivek Rao (2491, among top 60 in the US, top in Pennsylvania), receiving a provisional rating around 2570. ChipTest beat a 2354, drew a 2299, beat a 2421, drew a 2345 (?), beat Rao (2491) and lost to a 2321, receiving a provisional rating around 2501. Hitech had a rough outing, lost to a 2201 in the first round, beat and drew a few masters in the 2200 to 2360 range, and received a performance rating around 2312. BP did quite well for a micro. It finished with a respectable performance rating around 2189. If our calculation is correct, ChipTest should receive $100 for its performance. Along with the $2000 it won in the ACM, ChipTest has more than paid for its est. $500 cost (actually we never paid for the parts--they were leftovers from other projects). Both Deep Thought and ChipTest are definitely overrated at this moment. Vivek Rao, who lost to both programs, was probably overconfident. Before the game against ChipTest, he was openly expressing his contempt of chess playing computers (he had numerous successful encounters with Hitech earlier). ChipTest forced Rao to resign in under 30 moves with an unexpected sack. Vowing to take revenge for the loss on Deep Thought, but still expressing his contempt, he then proceeded to lose the last round game after Deep Thought played an unexpected pawn push that sent him into 25 minutes of deep thinking. If computer vs. computer rating does translate into computer vs. human rating, ChipTest should at best be 50 points above Hitech, or roughly 70 points below its provisional rating. We will probably never found out what ChipTest's real rating should be--this is ChipTest's last tourament. Both ChipTest and Deep Thought are authored by Thomas Anantharaman, Dr. Murray Campbell and yours truly of the Computer Science Department in Carnegie Mellon University. Some of ChipTest source code (under 0.5%, mainly in evaluation code) originated from Hitech, whose software development was headed by Dr. Hans Berliner of CMU with hardware designed by Dr. Carl Ebeling while he was at CMU. Deep Thought has its code completely rewritten, and does not contain any code from Hitech. Dr. Murray Campbell also worked on the Hitech project in association with Dr. Hans Berliner. Both Thomas and I are still graduate students (associated with the Speech group and the VLSI group respectively.). Deep Thought was still being wire-wrapped 2 days prior to the event. One point for the flakey hardware. I will post some of the games if there is interest. Article 1180 of rec.games.chess: Path: elroy!sun1!cracraft From: cracraft@sun1.uucp (Stuart Cracraft) Newsgroups: rec.games.chess Subject: Re: Computers in Fredkin Masters Open Keywords: Chess, Computers Message-ID: <6925@elroy.Jpl.Nasa.Gov> Date: 2 Jun 88 23:47:09 GMT References: <1828@pt.cs.cmu.edu> Sender: news@elroy.Jpl.Nasa.Gov Reply-To: cracraft@sun1.UUCP (Stuart Cracraft) Organization: Jet Propulsion Laboratory Lines: 21 Larry Kaufman has calculated the following ratings based on the Fredkin results: Chiptest 2496 (by method of CRA rating formula) 2504 (by method of linear formula) DeepThought 2588 (by method of CRA rating formula) 2586 (by method of linear formula) These values are based on a USCF-estimated-rating for Ivanov, the strong emigree whose FIDE rating was mentioned as 2415. USCF equivalent for this would probably be 2415+95 = 2510. Some future speculation: If DT's correct rating is USCF 2500, as is more likely -- with a good opening book and better endgame knowledge, a dual-processor version would probably be about USCF 2550. A full-fledged 100 processor version, would gain about 60-fold in speed, resulting in a USCF rating of about 2800, or FIDE 2700. So the program would come in just behind Kasparov and Karpov. Stuart ------------------------------ Date: Fri, 3 Jun 88 10:51 EDT From: Stephen G. Rowley Subject: References re AI in weather forecasting? Date: 26 May 88 15:53:01 GMT From: aplcen!jhunix!apl_aimh@mimsy.umd.edu (Marty Hall) Any pointers on where to look re AI in weather forecasting? I have a couple from AI in Engineering Proceedings, but can't find any others. Thanks! - Marty Hall -- ARPA - hall@bravo.cs.jhu.edu [hopkins-eecs-bravo.arpa] UUCP - ..seismo!umcp-cs!jhunix!apl_aimh | BITNET - apl_aimh@jhunix.bitnet Artificial Intelligence Laboratory, MS 100/601, AAI Corp, PO Box 126, Hunt Valley, MD 21030 (301) 683-6455 Well, I have 2 joke weather predictors for you: [1] The 1-rule version: Tomorrow's weather will be like today's weather. This is a pretty good predictor. Of course, you quickly find out people are interested in weather *changes*. :-) [2] The 4-rule version (New England only -- weather here is dominated by the ocean and cold Canadian air masses; we recently used this in a demo videotape about Joshua): * if the barometer is below 1000mbar and falling faster than a certain rate, o and if the temperature is below 40 . and if the wind is from the northwest, expect dry snow . else if the wind is from the northeast, expect wet snow . else if the wind is from the south, expect rain o else if the temperature is above 40, expect rain. This one is (slightly) less of a joke. I hooked it up to a Statice database of weather information gathered from a weather station at MIT. It usually manages to figure out it's raining about the time the raindrops hit my window. Now I get notifications on my lisp machine when it thinks it's raining... ------------------------------ Date: 29 May 88 13:22:48 GMT From: bpa!temvax!pacsbb!rkaplan@burdvax.prc.unisys.com (Randy Kaplan) Subject: Query on FRAME BASED LANGUAGES Frame Based Languages - Does anyone know of any more recent implementations of frame based languages, like FRL, that are currently available. I am doing research in knowledge acquisition and would like to use such a language as a representational platform. One written in Common LISP would be preferred. I am not interested in commercial packages unless they are both low cost and implement the notion of frames with languages like this, please let me know as soon as possible as we are in the midst of the research and would like to begin using the language as soon as possible. I can be contacted at, kaplan@vuvaxcom.bitnet. Thanks. Randy M. Kaplan Villanova University ------------------------------ Date: Fri, 3 Jun 88 18:22:57 EDT From: jesson@nrl-5570-gw (J.R. Jesson) Subject: References on parallel Inference Needed Hi! I'm in the beginning stages of developing a parallel inference engine. I've managed to gather some scattered references, but I haven't located many recent papers on this subject. I have Stanford, Texas and some CMU Tech reports up to early 1987, and access to many conference proceedings. If you have seen or authored papers concerning parallel inference, search, search heuristics, or the like, I would greatly appreciate hearing from you. Please drop me mail at one of the addresses below. Thanks... J.R. Jesson Merit Technology, Inc. 5068 W. Plano Parkway Plano, Texas 75075-5009 Voice: (214) 733-7092 UUCP : ...!ihnp4!killer!jesson ARPA : jesson@nrl-excalibur.arpa ------------------------------ Date: 3 Jun 88 11:51:36 GMT From: portal!cup.portal.com!Barry_A_Stevens@uunet.uu.net Subject: connectionist medical expert systems try contacting Dr. James A Anderson, Brown University, and ask about their database on antibiotics and diseases. Barry Stevens Presi2{ent {Applied AI Systems Del Mar CA 619-755-7231 ------------------------------ End of AIList Digest ********************